Explainable artificial intelligence in dental imaging: a systematic review of interpretability and the current state...

Explainable artificial intelligence (XAI) is increasingly used to improve the transparency of artificial intelligence models in dental and maxillofacial imaging, but the rigor with which explanations are evaluated remains unclear. This systematic review identified and categorized XAI approaches, their evaluation methods, and reported outcomes related to clinician trust and usability. The review was conducted in accordance with the PRISMA 2020 statement. Six electronic sources were searched, supplemented by Google Scholar and reference-list screening. Data were extracted on imaging modality, clinical application, explanation method, explanation scope, relationship to the predictive model, output format, evaluation approach, trust- or usability-related outcomes and external validation status. Risk of bias and applicability were assessed using QUADAS-2 and PROBAST. In total, 61 papers that met the specified inclusion criteria were identified. CAM-based local explanations predominated, whereas formal assessment of faithfulness, clinician trust, and usability was uncommon. Most included studies were judged to have a high overall risk of bias. Current evidence describes the technical use of XAI more strongly than its clinical utility. Future studies should use prespecified explanation taxonomies, quantitative faithfulness and localization tests, and clinician-centered evaluations of decision performance and calibrated reliance.

https://www.crd.york.ac.uk/PROSPERO/view/CRD420261421868.